AI Image Translation Model
2.4k 2026-04-18
GaParmar/img2img-turbo
A one-step image-to-image translation framework leveraging Stable Diffusion Turbo for rapid generation across various tasks like sketch-to-image and day-to-night transformations.
Core Features
Achieves one-step, highly efficient image-to-image inference.
Supports both paired (pix2pix-turbo) and unpaired (CycleGAN-Turbo) translation tasks.
Generates diverse outputs from a single input by varying noise maps.
Allows style control through text prompts.
Performs tasks such as sketch-to-image, edge-to-image, and day/night conversions.
Quick Start
python src/inference_paired.py --model_name "edge_to_image" --input_image "assets/examples/bird.png" --prompt "a blue bird" --output_dir "outputs"Detailed Introduction
img2img-turbo introduces a novel method for adapting single-step diffusion models like SD-Turbo to new image translation tasks through adversarial learning. It integrates three modules of latent diffusion models into a single end-to-end network, enabling efficient inference (e.g., 0.11s on A100 for 512x512 images). The project offers CycleGAN-Turbo for unpaired translation, outperforming existing GANs, and pix2pix-turbo for paired tasks, matching ControlNet's quality with superior speed. This framework significantly advances real-time image manipulation.